Security › hash-14 · Constant-time review · Performance: x86_1 · arm_1
Laurus (hash-14)
| Candidate | Laurus |
|---|---|
| Family | Symmetric (permutation-based) |
| Archive | Laurus.zip (SHA-256: 9499a772e532577f85902c0f631a95af8e2fb4bc308da6466c076debd63d1415) |
| Specification | hash-14-spec.pdf PDF |
| Performance | x86_1 · arm_1 |
hash-14-1: Laurus loses its function-domain separation at c=1024 and c=512
| Severity | Medium |
|---|---|
| Scope | design |
| Status | Confirmed |
| Affected | Specified generic Laurus[c,fid] interface at c=1024; named Laurus-512 and Laurus-XOF (c=512) for 1537-2560-bit messages; reference implementations |
| Discovery | Moderate |
| Exploitation | Trivial |
| Credit | Tsinghua Hash Lab cuihr26@mails.tsinghua.edu.cn (generic interface); named-function extension by Markku-Juhani O. Saarinen markku-juhani.saarinen@tuni.fi, with AI assistance |
| Date | 2026-09-22 |
Original source: CryptHashForum report Follow-up source: Laurus team's 2026-09-24 response
The specification uses fid=0 for hashing and fid=1 for XOF operation, placing the identifier only in the initial state. At c=1024, a message of 513 to 1024 bits causes one 512-bit absorption. That absorption omits the identifier from the permutation input and moves its difference into the first eight state words; finalization then discards exactly those words. The complete output is consequently independent of fid.
For example, the 768-bit all-zero message has identical 1024-bit outputs under Laurus[1024,0] and Laurus[1024,1]. This is a functional-domain separation failure, not a collision between distinct messages.
Extended on 2026-10-17: the same loss reaches the named functions. Laurus-512 is Laurus[512,0] and Laurus-XOF is Laurus[512,1] (§2.4-2.5, physical PDF page 11). The identifier sits only in the last state word (S_0 = [2^63 fid + c], §2.2, page 7; laurus_512.c:163). At c=512, g0 passes state words 8-23 to the feed-forward half without permuting them (:108-124). Each absorption therefore moves the identifier difference down eight words, from S_0[23] to S_1[15] to S_2[7], and g1 never reads S[7] (:127-141). Every message of 1537 to 2560 bits takes exactly two absorptions before finalization, so Laurus-512(M) = MSB_512(Laurus-XOF(M)) for every such M. For example, the 2048-bit all-zero message gives 309e429e…aad67432 under Laurus-512 and as the first 512 bits of Laurus-XOF. Shorter and longer messages remain separated. Through the uniform API, the two named functions are therefore related outputs rather than independent functions on this length window.
The Laurus team confirmed the generic-interface defect while reiterating that the named XOF uses c=512. As shown above, c=512 does not avoid the loss for 1537-2560-bit messages. The archived specification remains the target here.
Proposed fixes
The Laurus team's response proposes repeating the function-ID-bearing initialization value across all three 512-bit state segments. This section records the proposal without evaluating it.
Reproducing
Commands below run in a checkout of the ngcc-harness repository with the candidate built (see its README).
The target exposes the otherwise internal generic interface from the submitted source, checks the 768-bit equality, and checks a 512-bit negative control:
make -C hash-14 reproduce
The named-function extension uses the submitted Laurus-512 and Laurus-XOF libraries:
python3 hash-14/reproduce_named_xof_prefix.py
It tests every length from 0 to 3199 bits on three random streams and the all-zero stream. It requires equality for exactly the 1537-2560-bit lengths and none outside that window, and it checks the two named wrappers in the source (laurus_512.c:260, laurus_xof.c:261). It runs in under one second.
hash-14-2: Laurus-XOF ignores the requested output length and overruns shorter digest buffers
| Severity | Low |
|---|---|
| Scope | implementation |
| Status | Confirmed |
| Affected | Submitted Laurus-XOF reference CryptHash wrapper |
| Discovery | Trivial |
| Exploitation | A caller providing a 512- or 1024-bit output buffer receives a write beyond it; no secret disclosure or cryptographic break shown |
| Credit | fa1c4 (University of Science and Technology of China) |
| Date | 2026-10-20 |
Original source: NGCC harness issue 30
The XOF wrapper accepts digest_len_bits, but passes its fixed DIGEST_BIT_LENGTH = 2048 to laurus_xof instead (Laurus-XOF/CryptHash_AlgorithmInstance.c:14–18 and its header). A caller requesting 512 bits and allocating the corresponding 64 bytes is therefore written 256 bytes. The output-size argument is not checked or honored. The XOF core can take a variable output length; this is a wrapper error, not an attack on the hash construction.
Reproducing
Commands below run in a checkout of the ngcc-harness repository with the candidate built (see its README).
make -C hash-14 lib/libLaurus-XOF.so
python3 hash-14/reproduce_xof_output_length.py
The native call requests 512 bits in a larger sentinel-backed buffer, confirms that bytes past the requested 64 change, and compares the entire result with a 2048-bit control request. It does not make an out-of-bounds allocation or infer exploitability from the overwrite.